在AWS Lambda运行Inference SDK遇多进程错误，求解决方案

问题背景 #

在AWS Lambda上运行调用Roboflow Inference SDK的代码时，触发了多进程相关的OSError: [Errno 38] Function not implemented错误。代码未主动使用多进程，但依赖库底层调用了多进程组件，而AWS Lambda环境不支持多进程，已耗时9小时排查未解决，需要可行的解决思路。

运行代码 #

import os
from inference_sdk import InferenceHTTPClient


def handler(event, context):
    client = InferenceHTTPClient(api_url="https://detect.roboflow.com",
                                 api_key=os.environ["ROBOFLOW_API_KEY"])
    img_path = "./pizza.jpg"
    return client.infer(img_path, model_id="pizza-identifier/3")

Docker容器配置 #

FROM public.ecr.aws/lambda/python:3.11

RUN yum install -y mesa-libGL

COPY requirements.txt ${LAMBDA_TASK_ROOT}

RUN pip install -r requirements.txt

COPY pizza.jpg ${LAMBDA_TASK_ROOT}

COPY lambda_function.py ${LAMBDA_TASK_ROOT}

CMD [ "lambda_function.handler" ]

requirements.txt内容 #

inference==0.9.17

完整错误信息 #

{
  "errorMessage": "[Errno 38] Function not implemented",
  "errorType": "OSError",
  "requestId": "703be804-fd86-4b44-88f9-ac54c87717be",
  "stackTrace": [
    "  File \"/var/task/lambda_function.py\", line 10, in handler\n    return client.infer(img_path, model_id=\"pizza-identifier/3\")\n",
    "  File \"/var/lang/lib/python3.11/site-packages/inference_sdk/http/client.py\", line 82, in decorate\n    return function(*args, **kwargs)\n",
    "  File \"/var/lang/lib/python3.11/site-packages/inference_sdk/http/client.py\", line 237, in infer\n    return self.infer_from_api_v0(\n",
    "  File \"/var/lang/lib/python3.11/site-packages/inference_sdk/http/client.py\", line 299, in infer_from_api_v0\n    responses = execute_requests_packages(\n",
    "  File \"/var/lang/lib/python3.11/site-packages/inference_sdk/http/utils/executors.py\", line 42, in execute_requests_packages\n    responses = make_parallel_requests(\n",
    "  File \"/var/lang/lib/python3.11/site-packages/inference_sdk/http/utils/executors.py\", line 58, in make_parallel_requests\n    with ThreadPool(processes=workers) as pool:\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/pool.py\", line 930, in __init__\n    Pool.__init__(self, processes, initializer, initargs)\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/pool.py\", line 196, in __init__\n    self._change_notifier = self._ctx.SimpleQueue()\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/context.py\", line 113, in SimpleQueue\n    return SimpleQueue(ctx=self.get_context())\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/queues.py\", line 341, in __init__\n    self._rlock = ctx.Lock()\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/context.py\", line 68, in Lock\n    return Lock(ctx=self.get_context())\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/synchronize.py\", line 169, in __init__\n    SemLock.__init__(self, SEMAPHORE, 1, 1, ctx=ctx)\n",
    "  File \"/var/lang/lib/python3.11/multiprocessing/synchronize.py\", line 57, in __init__\n    sl = self._semlock = _multiprocessing.SemLock(\n"
  ]
}

解决思路 #

• 猴子补丁替换进程池为线程池：从栈跟踪可知，SDK误用了multiprocessing.pool.ThreadPool（实际基于进程池实现），可以用concurrent.futures.ThreadPoolExecutor替换该类，避免触发多进程逻辑。在lambda_function.py开头添加以下代码：

  import multiprocessing.pool
  from concurrent.futures import ThreadPoolExecutor
  
  def monkey_patch_threadpool():
      class ThreadPool:
          def __init__(self, processes, **kwargs):
              self.executor = ThreadPoolExecutor(max_workers=processes)
          def __enter__(self):
              return self.executor
          def __exit__(self, *args):
              self.executor.shutdown()
          def map(self, func, iterable):
              return list(self.executor.map(func, iterable))
      multiprocessing.pool.ThreadPool = ThreadPool
  
  monkey_patch_threadpool()
  

• 调整SDK版本：检查inference库的更新日志，尝试升级到最新稳定版或降级到0.9.17之前的版本，确认是否存在底层执行器的实现修复或变更。
• 直接调用Roboflow API：绕开SDK，手动构造HTTP请求调用检测接口，完全控制请求逻辑。示例代码：

  import os
  import requests
  
  def handler(event, context):
      api_key = os.environ["ROBOFLOW_API_KEY"]
      img_path = "./pizza.jpg"
      url = f"https://detect.roboflow.com/pizza-identifier/3?api_key={api_key}"
      with open(img_path, "rb") as f:
          response = requests.post(url, files={"file": f})
      return response.json()
  

• 排查运行时兼容性：尝试切换到Python 3.10的Lambda基础镜像，确认是否是Python 3.11与multiprocessing组件的兼容性问题；同时检查Lambda执行角色是否允许访问detect.roboflow.com的网络权限。

内容的提问来源于stack exchange，提问作者Dominique Paul